Applying the triads method in the validation
of dietary intake using biomarkers
Método das tríades na validação do consumo
alimentar com biomarcadores
1 Faculdade de Ciências da Saúde, Universidade de Brasília, Brasília, Brasil. 2 Departamento de Estatística, Universidade de Brasília, Brasília, Brasil.
Correspondence M. K. Ito
Faculdade de Ciências da Saúde, Universidade de Brasília.
Campus Darcy Ribeiro, Brasília, DF 70910-900, Brasil.
marina@unb.br
Renata Tiene de Carvalho Yokota 1 Edina Shizue Miyazaki 2
Marina Kiyomi Ito 1
Abstract
The triads method is applied in validation studies of dietary intake to evaluate the correlation be-tween three measurements (food frequency ques-tionnaire, reference method and biomarker) and the true intake using validity coefficients (ρ). The main advantage of this technique is the inclusion of the biomarker, which presents independent er-rors compared with those of the traditional meth-ods. The method assumes the linearity between the three measurements and the true intake and independence between the three measurement er-rors. Limitations of this technique include the oc-currence of ρ > 1, known as “Heywood case”, and the existence of negative correlations, which do not allow the calculation of ρ. The objective of this review is to present the concept of the method, de-scribe its application and examine the validation studies of dietary intake that use the triads meth-od. We also conceptualize the “bootstrap” method, used to estimate the confidence intervals of the validity coefficients.
Food Consumption; Nutrition Surveys; Validation Studies
Introduction
The food frequency questionnaire (FFQ) is the most widely used instrument for the assessment of habitual food intake of a population. A prope-rly validated FFQ for the intended population al-lows for stratification according to nutritional in-take at the reference time considered 1.
Biomark-ers offer the possibility of further validation aim-ing to improve the accuracy of the instrument 2.
In the process of validating a FFQ, multiple dietary records or 24-hour recalls (24hR) are of-ten used as the method of reference 3. The
ap-plication of a reference method is important in order to estimate the errors associated with the FFQ, particularly the attenuation bias caused by random measurement errors of the instrument which will impact in the statistical power of the study 1,3. The limitations of using dietary
mea-sures as the reference method is that both the instrument being tested (FFQ) and the reference method (24hR) are subject to the same random and systematic errors, because they rely on the memory of the interviewee and due to errors re-lated to the estimation of the reported food in-take 4,5.
In this scenario, biological markers may offer advantages and be able to improve the estimates of dietary intake assessment, due to the indepen-dence of their random errors in relation to the er-rors inherent to the intake questionnaires 5.
tradition-al methods of food intake 6. They should be used
as additional measures because not all nutrients have biological markers and many are influenced by factors other than intake, such as bioavail-ability, metabolism and genetic factors 5,7.
More-over, most of the biomarker analyses are expen-sive and quite often it is not possible to perform them as part of a large epidemiological study 6.
According to Kaaks 4, when information from
the FFQ, 24hR and biological markers measure-ments are available, the triangulation technique or the “method of triads” may be applied for the validation of the assessment methods of dietary intake. This method allows the comparison of food consumption estimated by the three meth-ods with the true (but unknown) intake by calcu-lating the validity coefficient (ρ) 4.
The aim of this review is to present the con-cept of the method, describe its implementation and examine the studies published from 1995 to 2009 in which the triangulation technique was applied in the validation of nutrient intake. An overview of the “bootstrap” resampling method for the estimation of the 95% confidence inter-vals (95%CI) of validity coefficients will be pre-sented, as well.
Method
Considering that the triads method is a relatively recent technique, we chose to extend the search to chapters in books, in addition to the papers published in specialized journals. Concepts, application procedures and validation stud-ies where triangulation method was used were sought. The “bootstrap” technique for the calcu-lation of confidence intervals of the validity coef-ficients is also reviewed.
We performed a literature search in the MEDLINE and PubMed databases for studies published in English, Spanish and Portuguese between 1995 and 2009, using a database search-ing filter and manual selection as needed. The keywords “triads method” or “validity coeffi-cient” was combined (AND) with the following terms (in parenthesis are, respectively, the num-ber of titles found with the combination “triads method” AND term; “validity coefficient” AND term): dietary intake (34; 252); dietary assess-ment (24; 139); biomarkers (52; 125). The com-bination of the keywords “triads method” AND “validity coefficient” resulted in 21 articles. The search retrieved 647 titles. After eliminating du-plicates, a total of 152 abstracts were identified. Using the final inclusion criteria which consisted of original articles that applied the triads meth-od, we found 16 studies, all of which were
in-cluded in this review. For the quality assessment and discussion of the selected articles, we used part of the criteria to evaluate dietary intake vali-dation studies (mainly the sample size criteria) proposed by Serra-Majem et al. 8. Although it
does not consider specific components used in the triads method, this is the first quality criteria tool developed for the assessment of validation studies of dietary questionnaires.
Results and discussion
Biomarkers of food intake
According to Kaaks et al. 7, the biomarkers of
die-tary intake can be classified into markers based on recovery or on concentration.
• Recovery-based markers are based on pre-cise and quantitative measures of the physiologi-cal balance between intake and excretion of a compound. Examples include 24-hour urinary nitrogen (for the intake of protein), urinary ex-cretion of potassium (for potassium intake) and doubly labeled water (for energy expenditure) 7.
Recovery markers have a direct and quantita-tive relation to nutrient intake. For instance, it is known that for any individual in protein and energy balance, the 24-hour urinary nitrogen represents approximately 80% of the nitrogen intake. On the assumption that only protein con-tributes significantly to dietary nitrogen content and its concentration in different types of protein is relatively constant, it is possible to estimate the absolute protein consumption of an individual from the quantity of nitrogen excreted in the 24-hour urine samples 7.
• Concentration-based markers are based on the concentration of a specific compound 7. The
concentration can be measured in biological ma-terials such as blood plasma or serum (carote-noids and tocopherols) 6,9,10, fractions of blood
li-pid (fatty acid in phospholili-pids) 11, adipose tissue
(carotenoids and tocopherols) 6 and urine
(nitro-gen) 6. These markers are not expressed in terms
of time units, and their quantitative relationship with intake may differ among individuals 7. Thus,
concentration-based biomarkers provide corre-lations with intake levels but can not be trans-formed into absolute measures of ingestion 7.
The triads method is being applied mostly in the validation studies that use concentration-based biomarkers.
Triads method: definition
Kaaks 4 proposed the triads method as a way to
quantitative intake information from the three methods (FFQ, 24hR and biological markers) was available. This method is an application of fac-tor analysis to this specific problem. The idea is that, although it is not possible to directly mea-sure the true intake (the latent variable), it can be estimated by means of FFQ and 24hR indicators, and biological markers, also known as manifest variables 4,12. The model assumes that the value
of each indicator can be decomposed into two components, one associated with the actual in-take and the other one to its own specificities. Mathematically, we can write:
FFQ = b10 + b11 | + e1;
24hR = b20 + b21 | + e2;
Biological markers = b30 + b31 | + e3.
Where, b denotes coefficients that relates | to FFQ, 24hR and biological markers; and e1, e2 and
e3 are the specific factors of each indicator. If the
factor of a particular indicator has little varia-tion, it means that this indicator provides a good approximation to the actual intake, ie, the corre-lation between the true intake and the indicator is high 4,12.
The assumptions for this technique are: the linearity of the relationship between the three variables and the true intake and independence of the specific factors 12. The assumption of
in-dependence implies that the correlations be-tween any pair of variables are due to the rela-tionship between each variable and the actual intake and not due to errors inherent in each assessment instrument (FFQ, 24hR and biologi-cal markers) 4,12,13.
The equations of this technique were gener-ated by the factor analysis model, although they can also be calculated by the structural equation analysis. Figure 1 illustrates the concept of the triads method triads 4,12.
The validity coefficients (ρ) are calculated by the following equations:
(1) ρQI =
(2) ρRI =
(3) ρBI =
Where, B = biological markers; Q = food fre-quency questionnaire (FFQ); R: 24-hour recalls (24hR).
From these equations, the correlation coef-ficients (r) between variables can be calculated:
(4) rQR = ρQI x ρRI;
(5) rQB = ρQI x ρBI;
(6) rBR = ρRI x ρBI.
Where, ρQI is the validity coefficient of FFQ
in relation to true intake, ρRI is the validity
coef-ficient of the reference method in relation to true intake, ρBI is the validity coefficient of the bio-marker in relation to true intake; rQR is the
cor-relation coefficient between the estimated intake by FFQ and reference method; rQB is the
corre-lation coefficient between the estimated intake by FFQ and biomarker and rBR is the correlation
coefficient between the estimated intake by ref-erence method and the biomarker 4,12.
It should be noted that the correlation coef-ficient used to calculate the validity coefcoef-ficient is the Pearson coefficient, when using numerical variables. The Spearman correlation coefficient can also be used when the interest is on the or-der of the variables 14. The validity coefficient for
the 3 variables (ρQI, ρRI and ρBI) can be calculated
using the formula described, with no need for a specific software 12.
The validity coefficients vary from 0 to 1, which is different from correlations coefficients, which range from -1 to 1. There are no negative validity coefficients, because this calculation in-cludes the square root. In general, the estimated validity coefficient for each variable (ρQI, ρRI and
ρBI) is equal to or greater than the correlations
between the variables (rQR, rRB and rBQ) 13.
When ρQI, ρRI and ρBI are high in relation to
the true intake, it is expected that the correlations between the manifest variables (rQR, rRB and rBQ)
are also relatively high. If the apparent correlation between two manifest variables is low, it suggests that at least one of the variables is not a good in-dicator of the true intake, resulting in low validity coefficient and wider confidence intervals 4,12.
The “bootstrap” method for calculating the 95%CI of the validity coefficient
In the estimation of validity coefficients, the ac-curacy of the coefficient may be evaluated from the confidence interval of the parameters.
The confidence interval of the validity coef-ficients are often calculated with the “bootstrap” method 6,10,15, originally proposed by Efron &
Gong 16. This is a resampling method in which
hundreds or thousands of “bootstrap” samples are generated from the original sample with the purpose of deriving estimates of confidence in-terval or standard errors of a parameter 14. Each
“boot-Figure 1
Triads method: triangular comparison between food frequency questionnaire, the reference method (24-hour recalls) and biological markers 4.
Q: food frequency questionnaire; R: reference method, B: biomarker; I: true intake; rQB: correlation between food frequency questionnaire and biomarker; rQR: correlation between food frequency questionnaire and reference method, rQR: correlation between biomarker and the reference method; ρQI: validity coeffi cient of food frequency questionnaire; ρRI: validity coeffi cient of the reference method; ρBI: validity coeffi cient of the biomarker.
strap” distribution of the validity coefficients, from which the confidence interval is calculated. This technique requires no prior knowledge of the estimated validity coefficient’s theoretical distri-bution 4. The confidence interval of the validity
coefficients can, then, be calculated by software programs such as SPSS (SPSS Inc., Chicago, USA) or Stata (Stata Corp., College Station, USA).
Thus, the “bootstrap” method provides an empirical distribution of the validity coefficients of the three (FFQ, 24hR and biological markers) variables. Large amplitudes of confidence inter-val will indicate low correlations between the variables 4.
Limitations of the triads method
One of the limitations of the triangulation tech-nique is the occurrence of validity coefficients
greater than one, known as the Heywood case 12.
For the three correlation coefficients (rQR, rRB and
rQB), the Heywood case occurs when the result of
the multiplication of two of the three correlation coefficients is greater than the other correlation coefficient for the same nutrient (eg.: rQR x rRB >
rQB) 12. For example, in the study by
Verkleij-Hagoort et al. 17, the validity coefficient between
FFQ and the “true intake” for vitamin B12 was
greater than 1 (ρQI = 1.66). Their results indicated
a Heywood case, where the product of rQR (0.66)
x rQB (0.21) was equal to 0.13, being larger than
rRB (0.05).
the result of systematic errors 4. Violation of the
assumption of independence of random errors between variables is more common, because FFQ and 24hR usually have correlated errors, particularly when 24hR or food records are used as reference methods. Therefore, some studies have considered the validity coefficient FFQ (ρQI)
as the upper limit and the correlation coefficient of FFQ and biological markers (rQB) as the lower
limit of the validity coefficient between FFQ and the true intake 10,17,18.
The existence of negative correlations for
rQR, rRB and rQB is another limitation of this
tech-nique, because validity coefficients (ρQI, ρRI and
ρBI) and ρ of the “bootstrap” samples cannot be
calculated 4,12. Empirical negative correlations
occur when the true correlations are near zero,
ie, the specific factors of the variable predomi-nate over the latent variable. High incidence of negative correlation means less precise confi-dence intervals which is due to the low accuracy of the estimated validity coefficients 4,12.
In-creasing the sample size and using more accu-rate reference methods and biomarkers should reduce the likelihood of negative correlations 4.
Therefore, in addition to using carefully chosen accurate reference methods, a minimum of 50 subjects is indicated for validation studies with biomarkers 8,15.
The validity coefficients of a nutrient are not always comparable between studies. Food con-sumption is culture specific, thus validity of di-etary intake measurement methods are estimated for different ethnic groups or study populations using reference methods for that population. Additionally, differences in the number of days of survey application, of the reference method used, sample size, the structure and number of food items of FFQ and the biomarker’s intrinsic variability (such as bioavailability and metabo-lism of the nutrient being tested) and analytical errors are some of the other factors that limit the comparability among studies 1,4.
Food consumption validation studies using the triads method
The nutrients investigated in the studies reviewed were: carotenoids (7) 6,9,10,15,18,19,20, tocopherols
(6) 6,9,10,15,18,20, retinol (1) 9, folic acid (4) 9,17,21,22,
fatty acids (4) 6,11,23,24, vitamin B12 (2) 17,22, proteins
(3) 9,15,25, potassium (2) 15,20, cholesterol (1) 20,
dithiocarbamate (1) 26, phytoestrogens (1) 27 and
flavonoids (1) 25. Table 1 presents a summary of
the studies in which the triangulation was used to analyze the correlation between methods of in-take, biomarkers and the unknown “true” intake. Nutrients most studied were carotenoids and
tocopherols. The majority of the studies aimed to validate a food intake questionnaire but some intended to test the biomarkers as indicators of food intakes 19,23,25,26.
The sample size of the studies varied from 27 to 161 individuals (Table 1). Recent guidelines by the European Micronutrient Recommendations Aligned Network of Excellence (EURRECA) con-sider satisfactory a sample size of 50 for validation studies with biomarkers as reference method 8.
Likewise, a sample of 100 individuals might be necessary for a validity coefficient with a 95%CI lower limit of at least 0.4, assuming correlation between FFQ and 24hR (rQR) of 0.6, 80% power
and 5% significance level 1. However, even this
sample size may often be insufficient to estimate validity coefficients precisely 4. Of the 16
pub-lished studies considered in this review, seven of them used sample sizes equal to or greater than 100 6,9,18,20,23,24,25, while four had sample sizes
be-low 50 10,11,21,26. Indeed, larger sample sizes will
reduce the chances of Heywood case events and the negative correlations between variables 9,12.
Validity coefficients greater than 1 and negative correlations were observed in the studies of vari-ous sample sizes reviewed. In one study, despite the moderate correlation (rQR = 0.57) obtained
between 24hR and FFQ for vitamin E, the corre-lation between biological markers and 24hR was negative (rRB = -0.11) 9. Small sample size (n = 28)
may account for this result. On the other hand, even with larger sample sizes, negative correla-tion was observed for adipose tissue α-carotene and palmitic acid 6. Considering the sample size
(n = 120) the Heywood cases were attributable to random errors due to low specificity of the biomarker 4. Therefore, sample sizes should
con-sider a biomarker with good quantitative relation to intake and the necessary study power, in addi-tion to the extra work imposed for the laboratory analyses.
Except for the dithiocarbamate 26, that used
urine as a sole biological material, 15 studies ana-lyzed biomarkers in blood samples or in adipose tissue (Table 1). For carotenoids, tocopherol, fo-lic acid and vitamin B12 studies, serum and
eryth-rocyte were analyzed. For fatty acid biomarkers analyses, plasma phospholipids 11, erythrocyte
membrane 24 and adipose tissue 6,23 were used.
In Kabagambe et al.’s 6 study, plasma performed
better than adipose tissue both for carotenoids and tocopherols. In the studies, blood collection was done in a fasting, fasting since midnight19
or non-fasting state 10,20. The effect of such
Table 1
Validation studies where validity coeffi cients (ρ) in relation to the “true” intake (I) were estimated for the triads method.
Reference Country n Dietary
intake: assessment
method
Biological sample
Bootstrap samples (n)
Validity coefficients (ρQI, ρRI, ρBI) and observations about the study
Carotenoids and tocopherol
Daurès et al. 15 (2000) France 87 adults 16 days WR 28 days WR
1 Q
Plasma 10,000 (0.39, 0.52, 0.85) for β-carotene. Q and WR were evaluated against B
Kabagambe et al. 6 (2001)
Costa Rica 120 adults 7 R24 2 Q
Plasma, adipose tissue
1,000 (0.45, 1.0, 0.36) for β-carotene (adipose tissue); (0.59, 0.81, 0.21) for α-tocopherol (adipose tissue) (0.76, 0.71, 0.50) for β-carotene in plasma Q was evaluated against 24R and two B. Overall plasma results were better than the adipose tissue ρ * Shai et al. 9 (2005) Israel 161 adults 6 R24
3 Q
Serum NI (0.67, 0.60, 0.67) for β-carotene (0.56, 0.97, 0.34) for α-tocopherol Performance of ρQI were superior to ρBI McNaughton et al. 10
(2005) **
Australia 28 adults 12 days WR 1 Q
Non-fasting serum
1,000 (0.55, 0.64, 0.51) for β-carotene
Not calculated for α-tocopherol (negative correlation). Q for carotenoid and vit E were evaluated against
24R and B Andersen et al. 19
(2005) ***
Norway 86-100 military men
14 days WR 2 Q (180
and 27 items)
Serum 1,000 (0.54, 0.79, 0.47) for 180 item Q and (0.60, 0.91, 0.43) for 27 item Q, both for α-carotene. Q were evaluated for fruit and vegetable
consumption Dixon et al. 18 (2006) # USA 130 adults 4 R24
1 DHQ
Serum (fasting from midnight)
NI (0.61, 0.42, 0.64) - β-carotene in women (0.80, 0.63, 0.71) - β-carotene in men
(0.74, 0.76, 0.14) for α-tocopherol DHQ was evaluated for carotenoid and tocopherol
intake. ρ were different according to gender and varied among nutrients tested Mirmiran et al. 20
(2010)
Iran 132 adults 12 R24 2 Q
Non-fasting plasma
NI (0.50, 0.48, 0.53) for β-carotene; (0.38, 0.39, 0.73) for α-tocopherol; (0.55, 0.51, 0.38) for retinol. Blood
samples were collected 4 times during a year Fatty acids
Kabagambe et al. 6 (2001)
Costa Rica 120 adults 7 R24 2 Q
Adipose tissue 1,000 (0.89, 0.82, 0.67) for 18:2n-6; (0.59, 0.99, 0.46) for 18:3n-3
Adipose tissue was a poor biomarker for saturated and monounsaturated fatty acids
Brevik et al. 23 (2005) ##
Norway 107-110 military men
14 days WR 1 Q
Serum and adipose tissue
1,000 (0.50, 0.94, 0.56) for 15:0 in adipose tissue and (0.58, 0.88, 0.49) for serum 15:0
Fatty acids 15:0 and 17:0 were tested as biomarkers of milk fat and dairy product intake. ρRI were superior
to others McNaughton et al. 11
(2007)
Australia 43 adults 12 days WR 1 Q
Plasma phospholipid
1,000 (0.45, 1.00, 0.29) for 20:4n-6; (0.62, 0.65, 0.35) for 20:5n-3; (0.62, 0.83, 0.52) for 22:6n-3
Zhang et al. 24 (2010)
China 125 adults 3DR 1 Q
Erythrocyte membrane
1,000 (0.61, 0.92, 0.17) for n-6; (0.65, 0.96, 0.29) for 18:3n-3; (0.75, 0.56, 0.50) for 20:5n-3; (0.67, 0.42,
0.24) for 22:6n-3
Folic acid and vitamin B12 Pufulete et al. 21 (2002) ###
United Kingdom
36 adults 7 day WR 2 Q
Serum and erythrocyte
NI (0.74, 0.72, 0.64) for serum folic acid (0.70, 0.78, 0.35) for erythrocyte folic acid. Validity coefficients were calculated using correlation values from the original
work Shai et al. 9 (2005) Israel 161 adults 6 R24
3 Q
Serum NI (0.72, 0.39, 0.65) for folic acid
Verkleij-Hagoort et al. 17 (2007)
Netherlands 53 women 3 R24 1 Q
Serum and erythrocyte
1,000 (0.94, 1.00, 0.21) for serum folic acid (0.75, 1.30, 0.37) for erythrocyte folic acid
(1.00, 0.39, 0.12) for serum vitamin B12 Q for folate and vitamin B12 was tested for women at reproductive age. ρRI, ρBI were calculated using
correlation values in the original work Shuaibi et al. 22 (2008) USA 95 women 3 day FR
1 FCM
Serum NI (0.97, 0.79, 0.46) for folic acid and (0.95, 0.85, 0.46) for Vitamin B12
Q for folate and vitamin B12 was tested in college-aged women
Other food components
Fowke et al. 26 (2002) USA 27 women 3 R24 2 FVQ
Urine 1,000 (0.54, 1.00, 0.42) before the intervention, (0.67, 0.49, 0.57) after intervention Urinary dithiocarbamate was analyzed as marker of cruciferous vegetable intake in an intervention study
aiming to increase its consumption
Bhakta et al. 27 (2005) United Kingdom
58 women 12 R24 1 Q
Plasma 10,000 (0.76, 0.82, 0.65) for genistein (0.67, 0.83, 0.45) for daizhein
Subjects were South Asian women living in England. Other 2 phytoestrogens were analyzed but results
were less consistent Shai et al. 9 (2005) Israel 161 adults 6 R24
3 Q
24-hour urine NI (0.77, 0.68, 0.44) urinary nitrogen for protein intake
Brantsaeter et al. 25 (2007)
Norway 119 pregnant
women
4 days WR 1 Q
24-hour urine, plasma
1,000 (0.59, 0.66, 0.47) for citrus fruit and juice, with zeaxanthin as biomarker,
(0.65, 0.63, 0.43) for citrus fruit and juice, considering Q plus hesperetin and zeaxanthin as biomarkers. Urine was used for flavonoid analyses and plasma for
carotenoids
Mirmiran et al. 20 (2010)
Iran 132 adults 12 R24 2 Q
24-hour urine, plasma
NI (0.56, 1.16, 0.38) for protein (0.61, 0.63, 0.60) for potassium (0.95, 0.46, 0.32) for cholesterol
Urine was used for nitrogen and potassium analyses. Plasma was used for cholesterol analysis
B: biological markers; DHQ: diet history questionnaire; n: sample size; FCM: “food choice map”; FR: food registry; FVQ: fruit and vegetable questionnaire; NI: not informed; Q: food frequency questionnaire; R24: 24-hours recall; WR: weighed record; ρQI: validity coeffi cient for food frequency questionnaire; ρRI: validity coeffi cient for reference intake method; ρBI: validity coeffi cient for biomarker.
* Mean validity coeffi cients for the tocopherols and carotenoids (α-carotene, β-carotene, lycopene, zeaxanthin and lutein); ** The study analyzed α and β-carotene, β-cryptoxanthin, lutein, lycopene;
*** The study analyzed lutein, zeaxanthin, lycopene, α and β-carotene;
# The study analyzed α and β-carotene, β-cryptoxanthin, lutein, lycopene, zeaxanthin, α and γ-tocopherol; ## Validity coeffi cient presented refers to dairy product as a whole;
### When validity coeffi cients were not available in the original work, they were calculated using Kaaks 4 equation. Table 1 (continued)
Reference Country n Dietary
intake: assessment
method
Biological sample
Bootstrap samples (n)
metabolism), are potential confounders for the biomarker and may result in weaker correlations to the true intake 1,4. The comparison between
studies must incorporate these important differ-ences among them.
For the intake reference method, studies used multiple 24hR 6,9,18,20,26,27, weighed food records 10,11,15,19,21,23 or several days food registry 22 as the
reference method. Taking the example of carote-noids and tocopherols, shown in Table 1, valid-ity coefficients of the reference method and the questionnaire performed well, ranging from ρRI
of 0.39 to 0.97 and ρQI from 0.38 to 0.80,
respec-tively. Some Heywood cases were also observed
6,11,16,17,26 for the reference method, and
poten-tial causes are violation of the assumption of in-dependence of variances 4,12. The biomarker ρ
BI
performed fairly well ranging from 0.36 to 0.71 for the carotenoids and 0.14 to 0.73 for tocopherols. Lower ρBI for tocopherol may indicate poor cor-relation with the latent variable or low specificity of the biomarker 4.
Dietary fatty acids are nutrients that are of validation interest from an epidemiological standpoint, due to their relationship with chron-ic diseases 28. Four studies applied the method
of triads 6,11,23,24, but comparison of the
stud-ies must be done with caution. The fatty acids were measured in different tissues (from blood and adipose tissue) and studies did not analyze same fatty acids. Overall, ρBI performance var-ied, ranging from 0.17 to 0.67 for essential fatty acids (18:2n-6 and 18:3n-3) 6,24 and odd number
fatty acids 23. Adipose tissue and serum showed
comparable results for the odd number fatty ac-ids but it was a poor biomarker for saturated and monounsaturated fatty acids 6. The validity
coef-ficients for the very long chain polyunsaturated fatty acids (20:5n-3 and 22:6n-3) varied among studies 11,24. Apparently, fatty acids that are not
synthesized in the body, such as the essential and odd number fatty acids, seems to perform better as biomarkers than those which are not derived solely from the diet (20:4n-6, 20:5n-3, 22:6n-3) 11.
This should be further investigated.
The method of triads was also used to com-pare different biomarkers for intakes of fruit and vegetables 19,25. In the study conducted in
Nor-way with male soldiers 19, the evaluation of
dif-ferent types of carotenoids (lutein, zeaxanthin, lycopene, a and b carotene) as biomarkers of fruit and vegetable intake showed α-carotene as having the best validity coefficient (ρBI = 0.47).
In addition, the authors compared a 180 item FFQ with a 27 item one, for the fruit and veg-etable consumption, using α-carotene as the biomarker. Both FFQ had similar validity coef-ficients (ρQI180 items = 0.54; ρQI27 items = 0.60),
showing that a FFQ with 27 items was sufficient to categorize fruit and vegetable intake in that population 19.
Overall, for folic acid and vitamin B12, the ρQI
and ρRI performed better than the ρBI (Table 1).
Still, these intake variables (FFQ and 24hR) have common sources of errors which violate the model assumption. Thus, this aspect needs to be taken in consideration when comparing the re-sults with that of the ρBI.
Biomarkers do not always perform better than other methods of food intake assessment. Among the 17 nutrients examined by Kabagam-be et al. 6, the α-tocopherol and β-carotene had
higher correlation coefficients between the FFQ and 24hR than between the biological markers (adipose tissue α-tocopherol and β-carotene) and FFQ (α-tocopherol rQB = 0.13; β-carotene
rQB = 0.16). Furthermore, the 95%CI for validity
coefficients of FFQ (ρQI = 0.12 to 1.00) and 24hR (ρRI = 0.40 to 1.00) were better than the biomarker
(ρBI = 0.02 to 0.67). These results indicate that for
some nutrients, the traditional methods of di-etary assessment are better than the biomarker, hence biomarkers should be used in addition to and not in replacement of dietary surveys 6.
Furthermore, the possibility of violation of the method’s assumption must be remembered for FFQ and 24hR.
Studies also used the method of triads to vali-date new biomarkers 23,26,27. One study tested
adipose and serum odd chain fatty acids as mark-ers of dairy fat intake 23 and another one used
urinary dithiocarbamates as biomarkers of cru-ciferous vegetable ingestion 26. In the
crucifer-ous study, authors conducted an intervention in which participants were encouraged to increase the consumption of cruciferous vegetables from 26.6g/day to 190.1g/day using a FFQ to estimate fruit and vegetable intake. They compared the consumption of these vegetables before and at the end of the intervention period using three 24hR as reference method and two fruit and vegetable intake questionnaires. Since the de-tection of urinary dithiocarbamate depends on the consumption of reasonable amount of cruciferous vegetables, the validity coefficient of the biomarker after intervention (ρBI = 0.65) was higher than the pre-intervention value (ρBI =
0.42). Although the authors recommend the use of this biomarker in the validation of cruciferous vegetable intake, the results need to be evaluated with caution, because the time reference of the FFQ was very short (7 days) and sample size dif-fered between before (n = 27) and after (n = 33) the intervention 26.
be-cause quite often validation studies are intended to relate food consumption over time with the development of chronic diseases. Bhakta et al. 27
conducted a study to validate a FFQ on phy-toestrogen intake by Asian women living in the United Kingdom (Table 1). Biomarkers used in the study were plasma phytoestrogens, which reflect their short term intake. Four plasma sam-ples were collected along a one year period. The biomarker validity coefficients (ρBI) were the
low-est for all phytolow-estrogens (0.11 for lignans and 0.45 for genistein). The results show that FFQ can be a better instrument than the biomarker in the estimation of phytoestrogen 27. Shai et al. 9 used
six repeated measures of 24hR, three FFQ, two blood samples and three urine samples, with the intent of obtaining a more precise measure of the attenuation factors for each variable stud-ied. Also, in the study conducted in Iran, four urine (potassium and protein intake estimate) and blood (β-carotene, α-tocopherol, retinol and cholesterol intake estimate) samples were col-lected 20. Urinary nitrogen 9,20 and potassium 20
performed well as biomarkers of protein and potassium intakes, respectively. These results should be useful in the adjustment of diet-dis-ease relative risk relationships in future studies.
The study by Brantsaeter et al. 25 used two
independent biomarkers (from 24-hour urinary flavonoids and plasma carotenoids) to validate a FFQ focused on fruits, vegetables and tea intake. Despite being costly and laborious, the inclusion of two nutrient biomarkers, one serving as the reference method, may show advantages due to the three independent variable errors generated. In their study, the highest validity coefficients
were seen for FFQ of citrus fruit/juice intake (ρQI = 0.65), using urine hesperetin and plasma
zeaxanthin as independent biomarkers. Com-parable results were obtained when the triads method was applied to two dietary estimates and one biomarker (ρQI = 0.59) 25.
Final considerations
The triads method is a technique that has been used in recent dietary validation studies. This method adds a third variable – the biomarker – with an independent error from FFQ and the reference method, allowing the expansion of the parameters for validation. The use of this meth-od does not exclude the need of correlation and Bland-Altman agreement analyses.
The number of published studies is still small and methodological differences related to popu-lation, the type of questionnaire and the refer-ence method hamper the comparability of the re-sults. Although most of the biomarkers behaved relatively well compared to the dietary estimates, the small sample size of some studies, together with other subject characteristics, like age, sex, supplement usage and smoking status (for car-otenoids) may have interfered in some results, which tended to be less correlated to the true in-take than the FFQ and 24hR methods.
Resumo
O método das tríades vem sendo utilizado em estudos de validação do consumo alimentar para avaliação da correlação entre três variáveis (questionário de freqüência alimentar, método de referência e biomar-cador) e a ingestão real, por meio dos coeficientes de validade (ρ). A principal vantagem deste método é a inclusão do biomarcador, que apresenta erros inde-pendentes dos métodos tradicionais. Os pressupostos desta técnica são a linearidade entre as três variáveis e a ingestão real, e a existência de erros independen-tes entre as variáveis. Entre as limitações deste méto-do, destaca-se a existência de ρ > 1, conhecido como “Heywood case”, e de correlações negativas, que não permitem o cálculo do ρ. O objetivo deste trabalho foi apresentar o conceito do método, descrever a sua apli-cação e examinar estudos de validação do consumo alimentar que utilizaram o método das tríades, além de conceituar o método “bootstrap” para obtenção de intervalos de confiança dos coeficientes de validade.
Consumo de Alimentos; Inquéritos Nutricionais; Estu-dos de Validação
Contributors
R. T. C. Yokota drafted the manuscript. E. S. Miyazaki contributed with the drafting and critical review of the statistical aspects of the article. M. K. Ito helped with the drafting and contributed with the critical review of the article.
References
1. Willett W, Lenart E. Reproducibility and validity of food frequency questionnaires. In: Willett W, edi-tor. Nutritional epidemiology. New York: Oxford University Press; 1998. p. 101-47.
2. Buzzard M. 24-hour dietary recall and food record methods. In: Willett W, editor. Nutritional epide-miology. New York: Oxford University Press; 1998. p. 50-73.
3. Gibson RS. Validity in dietary assessment methods. In: Gibson RS, editor. Principles of nutritional as-sessment. New York: Oxford University Press; 2005. p. 149-60.
4. Kaaks RJ. Biochemical markers as additional mea-surements in studies of the accuracy of dietary questionnaire measurements: conceptual issues. Am J Clin Nutr 1997; 65(4 Suppl):1232S-9S. 5. Potischman N. Biologic and methodologic issues
for nutritional biomarkers. J Nutr 2003; 133 Suppl 3:875S-80S.
6. Kabagambe EK, Baylin A, Allan DA, Siles X, Spiegelman D, Campos H. Application of the method of triads to evaluate the performance of food frequency questionnaires and biomarkers as indicators of long-term dietary intake. Am J Epide-miol 2001; 154:1126-35.
7. Kaaks R, Ferrari P, Ciampi A, Plummer M, Riboli E. Uses and limitations of statistical accounting for random error correlations, in the validation of di-etary questionnaire assessments. Public Health Nutr 2002; 5:969-76.
9. Shai I, Rosner BA, Shahar DR, Vardi H, Azrad AB, Kanfi A, et al. Dietary evaluation and attenuation of relative risk: multiple comparisons between blood and urinary biomarkers, food frequency, and 24-hour recall questionnaires: the DEARR Study. J Nutr 2005; 135:573-9.
10. McNaughton SA, Marks GC, Gaffney P, Williams G, Green A. Validation of a food-frequency ques-tionnaire assessment of carotenoid and vitamin E intake using weighed food records and plasma biomarkers: the method of triads model. Eur J Clin Nutr 2005; 59:211-8.
11. McNaughton SA, Hughes MC, Marks GC. Valida-tion of a FFQ to estimate the intake of PUFA using plasma phospholipid fatty acids and weighed food records. Br J Nutr 2007; 97:561-8.
12. Ocké MC, Kaaks RJ. Biochemical markers as ad-ditional measurements in dietary validity studies: application of the method of triads with examples from the European Prospective Investigation into Cancer and Nutrition. Am J Clin Nutr 1997; 65(4 Suppl):1240S-5S.
13. Kaaks R, Ferrari P. Dietary intake in epidemiology: can we know what we are measuring? Ann Epide-miol 2006; 16:377-80.
14. Hair Jr. JF, Anderson RE, Tatham RL, Black WC. Modelagem de equações estruturais. In: Hair Jr. JF, Anderson RE, Tatham RL, Black WC, organizado-res. Análise multivariada de dados. Porto Alegre: Bookman; 2005. p. 465-513.
15. Daurès JP, Gerber M, Scali J, Astre C, Bonifacj C, Kaaks R. Validation of a food-frequency question-naire using multiple-day records and biochemical markers: application of the triads method. J Epide-miol Biostat 2000; 5:109-15.
16. Efron B, Gong G. A leisurely look at the bootstrap, the jacknife and cross-validation. Am Stat 1983; 37:36-48.
17. Verkleij-Hagoort AC, Vries JHM, Stegers MPG, Lin-demans J, Ursem NTC, Steegers-Theunissen RPM. Validation of the assessment of folate and vita-min B12 intake in women of reproductive age: the method of triads. Eur J Clin Nutr 2007; 61:610-5. 18. Dixon LB, Subar AF, Wideroff L, Thompson FE,
Kahle LL, Potischman N. Carotenoid and tocoph-erol estimates from the NCI diet history question-naire are valid compared with multiple recalls and serum biomarkers. J Nutr 2006; 136:3054-61. 19. Andersen LF, Veierod MB, Johansson L, Sakhi A,
Solvoll K, Drevon CA. Evaluation of three dietary assessment methods and serum biomarkers as measures of fruit and vegetable intake, using the method of triads. Br J Nutr 2005; 93:519-27.
20. Mirmiran P, Esfahani FH, Mehrabi Y, Hedayati M, Azizi F. Reliability and relative validity of an FFQ for nutrients in the Tehran Lipid and Glucose Study. Public Health Nutr 2010; 13-654-62. 21. Pufulete M, Emery PW, Nelson M, Sanders AB.
Val-idation of a short food frequency questionnaire to assess folate intake. Br J Nutr 2002; 87:383-90. 22. Shuaibi AM, Sevenhuysen GP, House JD. Validation
of a food choice map with a 3-day food record and serum values to assess folate and vitamin B-12 in-take in college-aged women. J Am Diet Assoc 2008; 108:2041-50.
23. Brevik A, Veierod MB, Drevon CA, Andersen LF. Evaluation of the odd fatty acids 15:0 and 17:0 in serum and adipose tissue as markers of intake of milk and dairy fat. Eur J Clin Nutr 2005; 59:1417-22. 24. Zhang B, Wang P, Chen CG, He QQ, Zhuo SY, Chen
YM, et al. Validation of an FFQ to estimate the in-take of fatty acids using erythrocyte membrane fatty acids and multiple 3d dietary records. Public Health Nutr 2010; 13:1546-52.
25. Brantsaeter AL, Haugen M, Rasmussen SE, Alexan-der J, Samuelsen SO, Meltzer HM. Urine flavonoids and plasma carotenoids in validation of fruit, veg-etable and tea intake during pregnancy in the Nor-wegian Mother and Child Cohort Study (MoBa). Public Health Nutr 2007; 10:838-47.
26. Fowke JH, Hebert JR, Fahey JW. Urinary excretion of dithiocarbamates and self-reported cruciferous vegetable intake: application of the “method of tri-ads” to a food-specific biomarker. Public Health Nutr 2002; 5:791-9.
27. Bhakta D, Silva IS, Higgins C, Sevak L, Kassam-Khamis T, Mangtani P, et al. A semiquantitative food frequency questionnaire is a valid indicator of the usual intake of phytoestrogens by South Asian women in the UK relative to multiple 24-h recalls and multiple plasma samples. J Nutr 2005; 135:116-23.
28. Øverby NC, Serra-Majem L, Andersen LF. Dietary assessment methods on n-3 fatty acid intake: a sys-tematic review. Br J Nutr 2009; 102 Suppl 1:S56-63.
Submitted on 28/Aug/2009
